Correct binding table prefetch

Signed-off-by: Kamil Kopryk <kamil.kopryk@intel.com>
Related-To: NEO-6075

Binding table entry count was zeroed even when
ForceBtpPrefetchMode debug flag was enabled

level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_append_launch_kernel_3.cpp

@@ -32,6 +32,10 @@ HWCMDTEST_F(IGFX_GEN8_CORE, CommandListAppendLaunchKernel, givenFunctionWhenBind
     using MEDIA_INTERFACE_DESCRIPTOR_LOAD = typename FamilyType::MEDIA_INTERFACE_DESCRIPTOR_LOAD;
     using INTERFACE_DESCRIPTOR_DATA = typename FamilyType::INTERFACE_DESCRIPTOR_DATA;
 
+    for (auto debugKey : {-1, 0, 1}) {
+        DebugManagerStateRestore restore;
+        DebugManager.flags.ForceBtpPrefetchMode.set(debugKey);
+
         createKernel();
 
         ze_group_count_t groupCount{1, 1, 1};
@@ -61,6 +65,7 @@ HWCMDTEST_F(IGFX_GEN8_CORE, CommandListAppendLaunchKernel, givenFunctionWhenBind
             EXPECT_EQ(0u, idd->getBindingTableEntryCount());
         }
     }
+}
 
 HWCMDTEST_F(IGFX_GEN8_CORE, CommandListAppendLaunchKernel, givenEventsWhenAppendingKernelThenPostSyncToEventIsGenerated) {
     using GPGPU_WALKER = typename FamilyType::GPGPU_WALKER;

opencl/source/helpers/hardware_commands_helper_base.inl

@@ -251,10 +251,9 @@ size_t HardwareCommandsHelper<GfxFamily>::sendIndirectState(
 
     uint64_t offsetInterfaceDescriptor = offsetInterfaceDescriptorTable + interfaceDescriptorIndex * sizeof(INTERFACE_DESCRIPTOR_DATA);
 
-    auto bindingTablePrefetchSize = std::min(31u, static_cast<uint32_t>(kernel.getNumberOfBindingTableStates()));
+    auto bindingTablePrefetchSize = 0;
-
+    if (EncodeSurfaceState<GfxFamily>::doBindingTablePrefetch()) {
-    if (!EncodeSurfaceState<GfxFamily>::doBindingTablePrefetch()) {
+        bindingTablePrefetchSize = std::min(31u, static_cast<uint32_t>(kernel.getNumberOfBindingTableStates()));
-        bindingTablePrefetchSize = 0;
     }
 
     HardwareCommandsHelper<GfxFamily>::sendInterfaceDescriptorData(

shared/source/command_container/command_encoder.h

@@ -304,6 +304,7 @@ struct EncodeSurfaceState {
 
     static void getSshAlignedPointer(uintptr_t &ptr, size_t &offset);
     static bool doBindingTablePrefetch();
+    static bool isBindingTablePrefetchPreferred();
 
     static size_t pushBindingTableAndSurfaceStates(IndirectHeap &dstHeap, size_t bindingTableCount,
                                                    const void *srcKernelSsh, size_t srcKernelSshSize,

shared/source/command_container/command_encoder.inl

@@ -675,12 +675,18 @@ void EncodeIndirectParams<Family>::setWorkDimIndirect(CommandContainer &containe
     }
 }
 
+template <typename Family>
+bool EncodeSurfaceState<Family>::doBindingTablePrefetch() {
+    auto enableBindingTablePrefetech = isBindingTablePrefetchPreferred();
+    if (DebugManager.flags.ForceBtpPrefetchMode.get() != -1) {
+        enableBindingTablePrefetech = static_cast<bool>(DebugManager.flags.ForceBtpPrefetchMode.get());
+    }
+    return enableBindingTablePrefetech;
+}
+
 template <typename Family>
 void EncodeDispatchKernel<Family>::adjustBindingTablePrefetch(INTERFACE_DESCRIPTOR_DATA &interfaceDescriptor, uint32_t samplerCount, uint32_t bindingTableEntryCount) {
     auto enablePrefetch = EncodeSurfaceState<Family>::doBindingTablePrefetch();
-    if (DebugManager.flags.ForceBtpPrefetchMode.get() != -1) {
-        enablePrefetch = static_cast<bool>(DebugManager.flags.ForceBtpPrefetchMode.get());
-    }
 
     if (enablePrefetch) {
         interfaceDescriptor.setSamplerCount(static_cast<typename INTERFACE_DESCRIPTOR_DATA::SAMPLER_COUNT>((samplerCount + 3) / 4));

shared/source/command_container/command_encoder_bdw_and_later.inl

@@ -482,7 +482,7 @@ inline void EncodeSurfaceState<GfxFamily>::encodeExtraBufferParams(EncodeSurface
 }
 
 template <typename GfxFamily>
-bool EncodeSurfaceState<GfxFamily>::doBindingTablePrefetch() {
+bool EncodeSurfaceState<GfxFamily>::isBindingTablePrefetchPreferred() {
     return true;
 }
 

shared/source/command_container/command_encoder_xehp_and_later.inl

@@ -628,7 +628,7 @@ size_t EncodeMiFlushDW<Family>::getMiFlushDwWaSize() {
 }
 
 template <typename Family>
-bool EncodeSurfaceState<Family>::doBindingTablePrefetch() {
+bool EncodeSurfaceState<Family>::isBindingTablePrefetchPreferred() {
     return false;
 }
 

shared/source/gen11/command_encoder_gen11.cpp

@@ -21,7 +21,7 @@ using Family = NEO::Gen11Family;
 namespace NEO {
 
 template <>
-bool EncodeSurfaceState<Family>::doBindingTablePrefetch() {
+bool EncodeSurfaceState<Family>::isBindingTablePrefetchPreferred() {
     return false;
 }
 

shared/source/gen12lp/command_encoder_gen12lp.cpp

@@ -83,7 +83,7 @@ void EncodeSurfaceState<Family>::encodeExtraBufferParams(EncodeSurfaceStateArgs
 }
 
 template <>
-bool EncodeSurfaceState<Family>::doBindingTablePrefetch() {
+bool EncodeSurfaceState<Family>::isBindingTablePrefetchPreferred() {
     return false;
 }
 

shared/source/xe_hpc_core/command_encoder_xe_hpc_core.cpp

@@ -297,9 +297,6 @@ void EncodeDispatchKernel<Family>::appendAdditionalIDDFields(INTERFACE_DESCRIPTO
 template <>
 void EncodeDispatchKernel<Family>::adjustBindingTablePrefetch(INTERFACE_DESCRIPTOR_DATA &interfaceDescriptor, uint32_t samplerCount, uint32_t bindingTableEntryCount) {
     auto enablePrefetch = EncodeSurfaceState<Family>::doBindingTablePrefetch();
-    if (DebugManager.flags.ForceBtpPrefetchMode.get() != -1) {
-        enablePrefetch = static_cast<bool>(DebugManager.flags.ForceBtpPrefetchMode.get());
-    }
 
     if (enablePrefetch) {
         interfaceDescriptor.setBindingTableEntryCount(std::min(bindingTableEntryCount, 31u));